NotebookLM adds Audio Overviews

Background

NotebookLM works on the sources a user brings to it. Audio Overviews, added on 11 September 2024, applies that arrangement to sound rather than text: instead of a document to read, the user gets a conversation to listen to.^[1]

What launched

On 11 September 2024, Google added Audio Overviews to NotebookLM. From the Notebook guide, one click turns a user's sources into a "deep dive" — an audio discussion between two AI hosts who summarise the material and draw connections between topics.^[1]

The output is a dialogue, not a read-through. Two synthetic voices take turns, one picking up where the other left off. That is a different way of taking in a document: the listener can follow it while walking, commuting or doing something else, which is the practical point of putting notes into audio at all.

Google described the feature as experimental and named its limitations: the hosts could occasionally be inaccurate, there was no way for a listener to interrupt them, and the hosts spoke English only.^[1] The first two limitations point at the same property. Once generated, the audio was a fixed artefact — it could not be corrected, steered or questioned while playing.

Listening becomes a conversation

In December 2024, Google let users join an Audio Overview and put questions to the hosts.^[2] That changed the shape of the feature. The recording was no longer something a listener could only receive; joining the discussion and asking the two hosts a question made the audio a place to inquire rather than only to hear.

Video, and a life outside the product

Video Overviews followed in July 2025.^[2] The same underlying move — taking a user's sources and recasting them as a narrated, conversational summary — was extended from sound to picture.

The feature also turned up outside Google's own product. Spotify used NotebookLM to generate personalised AI podcasts in Spotify Wrapped 2024.^[2] Wrapped is Spotify's annual year-in-review for its listeners; a mainstream music service reaching for this tool to produce personalised audio put the two-host AI discussion format into a mass-market consumer product, not just an experimental corner of a research notebook.

Why it matters

For most of the history of AI assistants, the output was text on a screen, and the input was a question typed into a box. Audio Overviews reversed both habits: the input is the user's own pile of documents, and the output is speech, shaped as a conversation between two voices rather than a single narrator reading aloud. A dialogue has a natural structure — one host states something, the other picks it up, and the connection between two topics has to be said out loud rather than implied by layout. That makes the format suited to material that is long and loosely joined, where the useful information is spread across many pages.

The sequence of releases matters as much as the first version. The September 2024 feature was one-way and English-only and openly labelled experimental^[1]; the December 2024 change let listeners talk back to it^[2]; the July 2025 addition moved it from audio to video^[2]. By then the format had already travelled into a mainstream consumer campaign through Spotify Wrapped 2024.^[2] Listening to a summary is now one of the ordinary ways people get through their own material, alongside reading it.